"AI-Generated Image Detection Dataset v2 :10k+60k Paired Real and Synthetic Images for Benchmarking Detectors"
Abstract
"This dataset, AI-Generated Image Detection Dataset v2, provides a large-scale benchmark for distinguishing between real and AI-generated images. It contains 10,000 real images sourced from COCO and ImageNet-1k, paired with 60,000 synthetic images generated using six state-of-the-art text-to-image models: Stable Diffusion v1.5, SDXL, FLUX.1-schnell, Kandinsky 2.2, PixArt-Sigma, and Stable Cascade. Each real image is associated with six synthetic counterparts, all sharing a single caption generated by BLIP-2, ensuring consistent semantic alignment across pairs.All images are standardized to 512\u00d7512 resolution and processed through a canonical pipeline (EXIF-transpose, RGB conversion, cropping, resizing, JPEG equalization, and PNG export). The dataset is split deterministically into training (70%), validation (15%), and test (15%) subsets at the pair level, supporting reproducible experimentation. Captions are capped at 75 CLIP tokens to maintain compatibility with common vision-language models.The dataset is intended for non-commercial research use, inheriting licensing restrictions from ImageNet and COCO. It is particularly suited for developing and benchmarking detection models, studying generalization across generative architectures, and advancing research in computer vision and AI safety. Limitations include a focus on text-to-image generation (excluding img2img conditioning) and caption diversity bounded by BLIP-2 outputs. With approximately 24 GB of data, this…
Read the paper
Related studies
- Learning from First-Generation Qualitative Approaches in the IS Discipline: An Evolutionary View and Some Implications for Authors and Evaluators (PART 1/2)
- Commentary on the clinical and preclinical dosage limits of interstitially administered magnetic fluids for therapeutic hyperthermia based on current practice and efficacy models
- Prosthesis: New Technological Opportunities and Innovative Biomedical Devices
- The history and future of neural modeling for cochlear implants
- Ryanodine receptors: waking up from refractoriness
- Novel Transgenic Rabbit Model Sheds Light on the Puzzling Role of Matrix Metalloproteinase-12 in Atherosclerosis
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.